To determine annual patterns and correlates of nonfatal heroin overdose across 3 years, data were analyzed on 387 heroin users recruited for the Australian Treatment Outcome Study (ATOS), interviewed at 12, 24, and 36 months. A heroin overdose across follow-up was reported by 18.6%, and naloxone had been administered to 11.9%. Annual rates of overdose declined between baseline and 12 months and then remained stable. Previous overdose experience was strongly related to subsequent overdose. Those with a history of overdose before ATOS were significantly more likely to overdose during the study period. In particular, there was a strong association between overdose experience in any 1 year and increased overdose risk in the subsequent year. This is the first study to examine long-term annual trends in nonfatal heroin overdose. While overdose rates declined after extensive treatment, substantial proportions continued to overdose in each year, and this was strongly associated with overdose history.